Grant Hoffman was trained as a sketch artist before he became an AI solutions architect, and the most valuable lesson he learned during that time had nothing to do with drawing the subject in front of him. “If you are trying to depict something and it never seems to come out right, you draw the negative space around it,” he said during a From Day One webinar. “I don’t draw the camera, I draw the space around the camera, and by doing so, I draw the camera, because my brain has no concept of what the space around the camera is supposed to look like.”

Hofmann, now the AI solutions architect at Orange Logic, believes that lesson also applies to enterprise AI projects. Companies often point AI at the tasks it’s needed for, assuming the algorithm will figure out the rest, but the real work lies in mapping the invisible scaffolding around the task: brand rules, legal guardrails, and the tribal knowledge that only lives in the minds of a few seasoned employees. Without that layer of context, AI creates blur instead of drawing the camera.

That insight formed the core of the webinar’s conversation. Hoffman was joined by his colleague Misti Vogt, the SVP of engagement at Orange Logic, to lay out a five-step path for moving AI solutions from experiment to infrastructure, grounded in their work with organizations entrusted with managing millions of digital assets. 

Why Most AI Projects Disappoint

Hofmann says the biggest reason some AI initiatives fail is that companies treat AI as a magic bullet when it actually operates on the law of averages.

“These foundation models that most of us are building with are built on billions of parameters that represent relationships between a training data set that is mostly a single sum of human knowledge,” he said. “That is an average, and if we’re making assumptions based on the average, we are not capturing the nuance of how you or your organization really think about how to do a task.”

That nuance covers everything from tone of voice to rights management. “The way that I ask people to see it is as a magnifying glass,” Hofmann said. “If your process is good and it’s well thought out and it’s well documented, adding a little bit of AI to the mix is going to make it look 10 times more successful. But if your process is shaky, or if it relies on a lot of tribal knowledge, that’s where your hidden 10x can go the opposite direction.” Hofmann called these friction points “qualitative bottlenecks,” tasks where human intuition has always helped create the path. 

Vogt highlighted the foundational challenge organizations face when creating AI workflows: “The machines can’t be expected to know your brand rules, the governance for your brand, your voice, your rights, your taxonomy. So you have to create an ecosystem and infrastructure where they can easily lean into that information and get what they need when they need it.”

Vogt now calls this “enterprise content infrastructure.” Orange Logic has been building such platforms since 1998, and the company was recognized as a Leader by both Gartner and Forrester in 2025, she says.

From Record to Revenue

Vogt described a fundamental shift in content platforms. “Digital asset management has moved from really a system of record into a system of action, which puts us a lot closer to the revenue side of the business,” she said. “What you can do today is start benchmarking, because what’s happening for our customers who are starting to deploy these agents? It doesn’t translate really well, because they didn’t benchmark on the former process.”

Hofmann recommends identifying metrics that tie back to dollar-value outcomes: increased output, reduced legal review times, and lower tool bloat. “Cool factor doesn’t really hold water with the guy that signs the checks,” he added. “Before we fully get started, as we are in that scoping stage, we identify the metrics that we want to use in order to report, like, hey, this was really successful.” That discipline also prevents scope creep. “Being able to point to your metric for every new idea and say, 'How is this going to influence my metrics that I’m using to report value?’ Is this a good way to keep things on track?”

Teaching Machines Your Rules

Hofmann offered a low-tech exercise called the “sticky note method” to teach how to extract tribal knowledge within an organization. 

“Start with a pile of sticky notes next to you. Whether it’s you or whether you’re sitting down with your expert, we watch them, or we do the task ourselves. Every time our eyes start shifting to a different part of the screen, every time we consult a piece of prior knowledge, every single little piece of that gets documented by writing it down on the sticky note.” 

By the time the task is completed, the team has produced the first drafts of prompts and workflows, while highlighting steps that need a human touch. 

The method echoes some of the lessons Hofmann learned during his art training. “That sort of assumption engine that makes my day-to-day so fast and easy can get in my way or kill my accuracy,” he said. Vogt agreed: “It forces you out of common thinking. It’s not just fitting AI into current business processes; it’s rethinking the business processes.”

Crawl, Walk, Run

Hofmann and Vogt advocate for a phased approach to AI adoption that unfolds in three stages. The first, "crawl," focuses on using AI as a supportive tool, offering suggestions while humans guide and refine every action. In the "walk" phase, organizations introduce greater rigor by automating complete handoffs between steps and establishing clear benchmarks that make it easier to detect model drift, the point at which AI performance begins to decline. The final stage, "run," is reached when the system has been refined enough to consistently produce the desired output with minimal intervention.

Hofmann says that remaining in the crawl or walk phase is perfectly acceptable. “Run is a place that’s earned, not assumed.” Both speakers emphasized the importance of people remaining at the forefront of all AI processes. “Focus on people,” Hofmann said. “It’s not a replacement for humans; it should be an elevator for them.” 

Vogt recalled telling an employee who feared automation, “If you are able to automate your entire job with agents, you will become the most valuable employee that we have. So, go for it, push the limits, challenge it, test it.”

Hofmann ended the conversation with measured optimism. “The bad news is it’s harder than everyone thought it was going to be, but the good news is it’s also way easier than I think we think it is,” he said. “What I find is that AI projects are sort of this cascading explosion of success. It does not take a very long amount of time to go from our first successful AI project to starting to build an operating system that encompasses and enshrouds our business.” 

The secret to the success Hofmann has enjoyed is doing the human work first, drawing the negative space before you start drawing the camera. 

Editor’s note: From Day One thanks our partner, Orange Logic, for sponsoring this webinar. 

Ade Akin covers artificial intelligence, workplace wellness, HR trends, and digital health solutions.

(Photo by imaginima/iStock)